Abstract
We propose a concept of Virtual Kansei for robots and describe our attempts to construct a series of systems for generating virtual emotion, a subset of Virtual Kansei, of a robot. Here, Kansei is a Japanese word that encompasses affection, emotion, and related mental functions. Robots need to have human-like Kansei in order for them to become true partners of humans. The proposed virtual emotion consists of three modules; emotion detector to detect emotional state of the partner, emotion generator to generate robot emotion from partner’s emotion, and expression modulator to modify robot motions to express emotion. The emotion detector utilizes facial images, voice sounds, and body motions as inputs and Bayesian networks to integrate the information from three inputs. The emotion generator is a Petri-net that has a feedback structure to represent dynamics of emotion transition. Genetic Algorithm is adopted for adjusting arc weights of the Petri-net. The expression modulator uses an emotion vector to mix basis motions corresponding to the six emotions of Ekman. Experiments and results of three modules are discussed here.
